GB2562251A

GB2562251A - System and method for detecting unauthorised personnel

Info

Publication number: GB2562251A
Application number: GB1707414.7A
Authority: GB
Inventors: Hunt Brian
Original assignee: Zircon Software Ltd
Current assignee: Zircon Software Ltd
Priority date: 2017-05-09
Filing date: 2017-05-09
Publication date: 2018-11-14
Also published as: GB201707414D0

Abstract

A system or method suitable for determining whether a person in the vicinity of a railway track is authorised or unauthorised, the system or method comprising receiving an image in the vicinity of a railway track, identifying a person in the image and identifying whether the person is wearing authorised clothing. An optical detector may capture colour and/or infra-red images. Identifying a person in the image may comprise a processor configured to calculate feature descriptors of the image including histogram of orientated gradients descriptors 22 and configured to apply a cascade classifier 26 method to the feature descriptors. The system/method may comprise determining overlapping detected persons and discarding the smaller by non-maxima suppression 28. The system/method may include a colour/light filter 30/32 to identify whether the person is wearing a garment of a predetermined colour/light level wherein the image may be a colour/greyscale image respectively. An alert may be output if an unauthorised person is detected and a timestamp and/or geographical position may be associated with the image.

Description

System and method for detecting unauthorised personnel

TECHNICAL FIELD

The present invention relates to a system and method for detecting unauthorised personnel. In particular the present invention relates to a system and method for detecting unauthorised personnel in the vicinity of a railway line.

BACKGROUND

The safety of a railway network is very important and one hazard is trespassers on the railway lines. Trespassers on the railway lines run the risk of collision with trains and electrocution from electric overhead lines and electrified rails. Railway lines are typically protected by barriers, such as fencing, to prevent pedestrians from accessing the lines. However, there remain instances of trespassers getting onto the lines and endangering themselves and others.

It is routine to have maintenance work on railway network, which requires authorised personnel in the vicinity of the railway lines. Authorised personnel wear standard high visibility (Hi-Vis) clothing. For example, authorised personnel working for Network Rail™, the infrastructure manager for most of the rail network in England, Scotland and Wales wear a standard fluorescent orange with retroreflective strips.

It would be advantageous to have improved detection of trespassers on the railway line, to allow oncoming trains to slow down or stop and also to identify areas where security around the lines can be improved.

SUMMARY OF THE INVENTION A first aspect of the invention provides a system for determining whether a person in the vicinity of a railway track is authorised or unauthorised, the system comprising: means for receiving an image of the vicinity of a railway track; means for identifying a person in the image; and means for identifying whether the person is wearing authorised clothing. A processor may be configured to receive an image of the vicinity of a railway track, identify a person in the image and identify whether the person is wearing authorised clothing. The processor may comprise a dedicated processor or may comprise a processor on a computer, such as a PC.

The image may comprise a recorded or live image. The system may comprise an optical detector configured to capture images of the vicinity of a railway track. The optical detector may comprise a camera, for example a digital camera or CCTV. The optical detector preferably produces a pixelated image. The optical detector may produce colour images during the daytime. The optical detector may produce a greyscale image at night. The optical detector may have a frame rate of >60 frames per second. This frame rate enables the system to be used with high speed vehicles, such as high speed trains. A lower frame rate may be used with vehicles travelling at lower speeds.

The optical detector may be mounted on a train. Alternatively, the optical detector may be statically mounted relative to the railway track. The optical detector may be configured to capture colour images. The optical detector may be configured to capture infra-red illuminated images.

The means for identifying a person in the image may comprise a processor configured to calculate feature descriptors ofthe image. The feature descriptors may comprise histogram of oriented gradients descriptors. The processor may be configured to apply a classifier method to the calculated feature descriptors to determine a detected person in the image. The classifier method may comprise a cascade classifier method.

The processor may be configured to determine overlapping detected persons and discard the smaller of the overlapping detected persons. The step of determining overlapping detected persons may comprise non-maxima suppression.

The means for identifying whether the person is wearing authorised clothing may comprise a filter to determine whether the image relating to the identified person is wearing a garment of a predetermined colour and/or predetermined light level. The image may be a colour image and the filter may comprise a colour filter. The image may be a greyscale image and the filter may comprise a light filter. The means for identifying whether the person is wearing authorised clothing may comprise a processor configured to determine whether a section of the image identified as a person includes a percentage of a predetermined colour above a predetermined threshold. The percentage may comprise about 90%, 80%, 70%, 60%, 50%, 40% or 30%. The means for identifying whether the person is wearing authorised clothing may comprise a processor configured to determine whether a section of the image identified as a person includes a percentage of a predetermined light level above a predetermined threshold. The percentage may comprise about 90%, 80%, 70%, 60%, 50%, 40%, 30%, 20% or 10%. The means for identifying whether the person is wearing authorised clothing may comprise a processor configured to determine the area within a boundary ofthe person identified in the image; calculate the percentage of that area which has the predetermined colour and/or light level; and compare this percentage against a threshold percentage. The image may comprise multiple pixels and wherein determining the area within a boundary may comprise counting the number of pixels within that boundary.

The system may comprise means for outputting an alert if an unauthorised person is detected, for example an audible alarm, email message or SMS message.

The system may comprise a memory for saving a detected image. The memory may save a timestamp and/or geographical position associated with the detected image. The system may comprise a geolocation device, such as a GPS, to generate a geographical position associated with image. The system may comprise a clock to timestamp the image. A second aspect of the invention provides a method for determining whether a person in the vicinity of a railway track is authorised or unauthorised, the method comprising: receiving an image of the vicinity of a railway track; identifying a person in the image; and identifying whether the person is wearing authorised clothing.

The step of identifying a person in the image may comprise the step of calculating feature descriptors of the image. The feature descriptors may comprise histogram of oriented gradients descriptors. The step of identifying a person in the image may comprise the step of applying a classifier method to the calculated feature descriptors to determine a detected person in the image. The classifier method may comprise a cascade classifier method. The method may comprise the step of determining overlapping detected persons and discarding the smaller overlapping detected person. The step of determining overlapping detected persons and discarding the smaller comprises non-maxima suppression.

The step of identifying whether the person is wearing authorised clothing may comprise applying a filter to determine whether the image relating to the identified person is wearing a garment of a predetermined colour and/or predetermined light level. The image may be a colour image and the filter may comprise a colour filter. The image may be a greyscale image and the filter may comprise a light filter. The step of identifying whether the person is wearing authorised clothing may comprise determining whether a section of the image identified as a person includes a percentage of a predetermined colour above a predetermined threshold. The step of identifying whether the person is wearing authorised clothing may comprise determining whether a section of the image identified as a person includes a percentage of a predetermined light level above a predetermined threshold. The step of identifying whether the person is wearing authorised clothing may comprise determining the area within a boundary of the person identified in the image; calculating the percentage of that area which has the predetermined colour and/or light level; and comparing this percentage against a threshold percentage. The image may comprise multiple pixels and wherein determining the area within a boundary may comprise counting the number of pixels within that boundary.

The method may comprise outputting an alert if an unauthorised person is detected. The method may comprise saving a detected image in a memory. The method may comprise saving a timestamp and/or geographical position associated with the detected image.

The present invention is not limited to identifying authorised/non-authorised personnel in the vicinity of a railway track. The present invention is also suitable for identifying authorised/non-authorised personnel in other areas in which authorised personnel wear standard clothing. This is particularly important in areas where non-authorised personnel can impact on safety, such as around roads and in emergency situations. The invention is particularly suitable for differentiating authorised personnel such as road workers, emergency services and security guards from unauthorised personnel. In these cases, the optical detectors may comprise CCTV cameras or may comprise optical detectors mounted on vehicles, such as cars. A third aspect of the invention provides a system for determining whether a person in an area is authorised or unauthorised, the system comprising: an electronic processor having an electrical input for receiving an image of the area, the image being automatically generated and output by a camera; an electronic memory device electrically coupled to the electronic processor and having instructions stored therein; wherein the electronic processor is configured to access the memory device and execute the instructions stored therein such that it is operable to: identify a person in the image; and identify whether the person is wearing authorised clothing. A fourth aspect of the invention provides a system for determining whether a person in an area is authorised or unauthorised, the system comprising: means for receiving an image of the area; means for identifying a person in the image; and means for identifying whether the person is wearing authorised clothing. A fifth aspect of the invention provides a method for determining whether a person in an area is authorised or unauthorised, the system comprising: receiving an image of the area; identifying a person in the image; and identifying whether the person is wearing authorised clothing.

Throughout the description and claims of this specification, the words “comprise” and “contain” and variations of the words, for example “comprising” and “comprises”, mean “including but not limited to”, and do not exclude other components, integers or steps. Moreover the singular encompasses the plural unless the context otherwise requires: in particular, where the indefinite article is used, the specification is to be understood as contemplating plurality as well as singularity, unless the context requires otherwise.

Preferred features of each aspect of the invention may be as described in connection with any of the other aspects. Within the scope of this application it is expressly intended that the various aspects, embodiments, examples and alternatives set out in the preceding paragraphs, in the claims and/or in the following description and drawings, and in particular the individual features thereof, may be taken independently or in any combination. That is, all embodiments and/or features of any embodiment can be combined in any way and/or combination, unless such features are incompatible.

BRIEF DESCRIPTION OF THE DRAWINGS

One or more embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings, in which:

Figure 1 is a schematic illustration of the main components of an embodiment of the invention; and

Figure 2 is a schematic illustration of the processing steps of an embodiment of the invention.

DETAILED DESCRIPTION A schematic illustration of the main components according to an embodiment of the invention is shown in Figure 1. An optical detector comprising a camera 10 is mounted in a position to capture images of the vicinity of a railway track, i.e. the track itself and the region surrounding it. The camera 10 may be mounted on a train, for example on the front of a train pointing forwards or to the rear of a train pointing backwards. Alternatively, the camera 10 may be statically mounted, for example a statically mounted CCTV camera. A geolocation device 12, such as a global positioning system (GPS) is optionally provided which allows a geographical location to be provided for each image. The system may comprise a timer (not shown) to timestamp both the images from the camera 10 and the geographical location data from the geolocation device 12. A memory 14 is provided for storing the images from the camera 10 and optionally geographical coordinates from the geolocation device 14 for later offline processing. For processing from live footage from the camera, the images and optionally the geographical location data are sent to processor 16 which analyses the images and determines from the image whether unauthorised personnel are on the railway line. If unauthorised personnel are detected, an event notification 18 is issued.

The processor may comprise any processor suitable for image processing, for example a dedicated processor or a personal computer.

In the processor, images are analysed to detect people within the image. The system is able to take each detection of a person, determine whether the detection relates to authorised or unauthorised personnel, and then raise an alert in the latter case. An authorised person is one who is wearing authorised clothing, for example an orange Hi-Vis jacket.

The accuracy of the detection of a person from the image is dependent upon the resolution of the images, the contrast of the image, and the capture framerate. The system should achieve or exceed 60 frames per second in order to be able to function effectively on trains travelling at their maximum line speed (125mph = 48 meters/second in the UK).

Colour images are captured during the day and greyscale images are captured at night. The camera is capable of capturing IR illuminated images as well as colour images, which enables it to detect retroreflective strips on clothing at night, as well as the orange colour in the daytime. The framerate of the capture (and therefore responsiveness of the detection) is dependent on the processing capacity of the system. Unless recording footage from the cameras, the device requires minimal physical maintenance.

Figure 2 is a schematic diagram illustrating the processing steps of the system. The system follows a simple processing loop, repeating the listed steps for as long as the system is operational. These stages must be processed as quickly as possible, as delays in one stage will slow down all other stages and reduce the accuracy of the system.

Frame capture 20

In order to begin the process, an image is required to perform the process upon. While the process described here assumes real time processing, it is also possible to pre-record a video from the camera and then perform the analysis later.

The camera is mounted facing outwards from the front (or rear) of the train. The camera needs to be placed with minimum obstructions, and as wide angle view as is possible.

Frame capture is the first stage in the processing chain. It runs continuously, as fast as computationally possible, passing each frame to a processing queue for processing.

The camera is likely to be connected to the processor via FireWire, USB3, or Ethernet.

Histogram of oriented gradients 22

Before detection of unauthorised personnel is possible, the frame needs to be formatted into a form the processor can analyse, by calculating the feature descriptors of the image.

The system uses Histogram of Orientated Gradients (HOG) descriptors. HOG calculates the gradient values from across the image, orientates them via cell histograms, and groups the cells into larger descriptor blocks that can be then used to analyse the image and make decisions as to the location of persons.

Classifier Detection 24

The classifier takes the computed feature descriptors and uses them to make decisions about the presence of a person at each location.

This embodiment uses the cascade classifier method. This method uses multiple weak and computationally cheap detections of persons within an image, and combines the outputs to result in much stronger detection. Further, these classifiers can be trained (effectively calibrating the weak detectors) multiple times in order to refine the strength of the detection further. The Classifier XML (Extensible Mark-up Language) 26 is the result of this training; a list of the setting each weak detector should use for optimum performance which is used to initialise a trained classifier.

The system’s classifier has been trained to detect the upper body (head and torso) of humans, using HOG descriptors. The classifier returns a list of detections, which are then passed on to the non-maxima suppression.

Non-maxima suppression 28

Each detection returned by the classifier is in the form of [Upper left co-ordinate, Bottom right co-ordinate]. First, this list of detections is filtered for overlapping values, using non-maxima suppression. This simply compares two overlapping detections and discards the smaller, until there are no longer any overlapping regions.

Both the frame and the filtered list of detections are then fed into either the colour filter, if it is a daylight image with colour, or the light filter, if it is a greyscale night-time image.

Colour filter 30/light filter 32

The filters, both colour and light, search the detections for signs that the person detected is authorised personnel. It is assumed that all authorised personnel will be wearing authorised Hi-Vis clothing.

Colour filtering 30 cuts out the detection area from the frame, and then searches for the presence of a Hi-Vis clothing via the colour of the garment. It counts the number of pixels within the defined colour boundary, calculates that as a percentage of the entire detection, and compares that percentage against a threshold percentage. The threshold percentage is a configurable value to prevent small amounts of the authorised colour registering the person as authorised; a substantially full body covering is required.

Light filtering 32 cuts out the detection area from the frame, and then searches for the presence of a Hi-Vis via the brightness of the garment’s reflective surfaces. It counts the number of pixels above a defined brightness boundary, calculates that as a percentage of the entire detection, and compares that percentage against the threshold percentage.

Regardless of how the detections are filtered, they are separated into two categories; Authorised and Non-authorised.

Results 34

The system may be integrated with other systems to provide alerts to operators, for example audible alarms or email messages.

Alternatively, the lists of authorised and non-authorised detections can be displayed to the operator by drawing coloured boxed around the detections on a dedicated user interface. Authorised personnel are surrounded by a dark green box, and unauthorised pedestrians are surrounded by a bright red box.

The detection image frames can then be saved with a timestamp and geographical position to a file or database for evidence purposes.

This invention can be used to alert the train driver of the presence of unauthorised personnel on the railway line so that the driver can slow or stop the train. In addition, the invention can be used to collect information of areas where trespassing occurs, so action can be taken to improve rail security.

Although this embodiment describes authorised personnel wearing standard orange Hi-Vis jackets, the system and method can be configured for use with other types of authorised clothing, for example different colours or without retroreflective strips.

Claims

1. A system for determining whether a person in the vicinity of a railway track is authorised or unauthorised, the system comprising: means for receiving an image of the vicinity of a railway track; means for identifying a person in the image; and means for identifying whether the person is wearing authorised clothing.

2. A system according to any preceding claim, wherein the image comprises a recorded or live image.

3. A system according to any preceding claim, comprising an optical detector configured to capture images of the railway track.

4. A system according to claim 3, wherein the optical detector has a frame rate of >60 frames per second.

5. A system according to any of claims 3 or 4, wherein the optical detector is mounted on a train.

6. A system according to any of claim 3 or 4, wherein the optically detector is statically mounted relative to the railway track.

7. A system according to any of claims 3 to 6, wherein the optical detector is configured to capture colour images.

8. A system according to any of claims 3 to 7, wherein the optical detector is configured to capture infra-red illuminated images.

9. A system according to any preceding claim, wherein the means for identifying a person in the image comprises a processor configured to calculate feature descriptors of the image.

10. A system according to claim 9, wherein the feature descriptors comprise histogram of oriented gradients descriptors.

11. A system according to any of claims 9 or 10, wherein the processor is configured to apply a classifier method to the calculated feature descriptors to determine a detected person in the image.

12. A system according to claim 11, wherein the processor is configured to determine overlapping detected persons and discard the smaller.

13. A system according to any preceding claim, wherein the means for identifying whether the person is wearing authorised clothing comprises a filter to determine whether the image relating to the identified person is wearing a garment of a predetermined colour and/or predetermined light level.

14. A system according to claim 13, wherein the image is a colour image and the filter comprises a colour filter.

15. A system according to claim 13, wherein the image is a greyscale image and the filter comprises a light filter.

16. A system according to any preceding claims, wherein the means for identifying whether the person is wearing authorised clothing comprises a processor configured to determine whether a section of the image identified as a person includes a percentage of a predetermined colour above a predetermined threshold.

17. A system according to any of claims 1 to 15, wherein the means for identifying whether the person is wearing authorised clothing comprises a processor configured to determine whether a section of the image identified as a person includes a percentage of a predetermined light level above a predetermined threshold.

18. A system according to any preceding claims, comprising means for outputting an alert if an unauthorised person is detected.

19. A system according to any preceding claims, comprising a memory for saving a detected image.

20. A method for determining whether a person in the vicinity of a railway track is authorised or unauthorised, the system comprising: receiving an image of the vicinity of a railway track; identifying a person in the image; and identifying whether the person is wearing authorised clothing.

21. A method according to claim 20, wherein identifying a person in the image comprises the step of calculating feature descriptors of the image.

22. A method according to claim 21, wherein the feature descriptors comprise histogram of oriented gradients descriptors.

23. A method according to any of claims 21 to 22, wherein identifying a person in the image comprises the step of applying a classifier method to the calculated feature descriptors to determine a detected person in the image.

24. A method according to claim 23, wherein the classifier method comprises a cascade classifier method.

25. A method according to claim 24, comprising the step of determining overlapping detected persons and discarding the smaller.

26. A method according to claim 25, wherein the step of determining overlapping detected persons and discarding the smaller comprises non-maxima suppression.

27. A method according to any of claims 20 to 26, wherein identifying whether the person is wearing authorised clothing comprises applying a filter to determine whether the image relating to the identified person is wearing a garment of a predetermined colour and/or predetermined light level.

28. A method according to claim 27, wherein the image is a colour image and the filter comprises a colour filter.

29. A method according to claim 13, wherein the image is a greyscale image and the filter comprises a light filter.

30. A method according to any of claims 20 to 29, wherein identifying whether the person is wearing authorised clothing comprises determining whether a section of the image identified as a person includes a percentage of a predetermined colour above a predetermined threshold.

31. A method according to any of claims 20 to 29, wherein identifying whether the person is wearing authorised clothing comprises determining whether a section of the image identified as a person includes a percentage of a predetermined light level above a predetermined threshold.

32. A method according to any of claims 20 to 29, wherein identifying whether the person is wearing authorised clothing comprises determining the area within a boundary of the person identified in the image; calculating the percentage of that area which has the predetermined colour and/or light level; and comparing this percentage against a threshold percentage.

33. A method according to claim 32 wherein the image comprises multiple pixels and wherein determining the area within a boundary comprises counting the number of pixels within that boundary.

34. A method according to any of claims 20 to 33, comprising outputting an alert if an unauthorised person is detected.

35. A method according to any of claims 20 to 34, comprising saving a detected image in a memory.

36. A method according to claim 34, comprising saving a timestamp and/or geographical position associated with the detected image.